I am shortlisting my candidates for validation using qPCR/RTPCR of my results from two different splicing pipelines(DEXSeq & Cufflinks/Cuffdiff). However, I am a little lost as to what candidates I should pick to validate alternative splicing between two conditions/samples. I have picked up candidate exons for validation from DEXSeq results. However, just validating at the exon level seems insufficient to me and I want to validate at the transcript level. I was analyzing potential candidates from my cuffdiff results to find interesting transcripts. However, I am having trouble finding good candidates.
Here's my approach:
Fist I looked at the results from splicing.diff and identified genes that show statistically significant alternative splicing (based on pvalue of the JS metric). From here I identified the tss_id of the candidate gene (picked one to start if multiple present). Using this tss_id, I went on to identify the transcripts that have this tss_id (in theory spliced from sample primary transcript or promoter). Then, I checked the tss_id in tss_group_exp.diff file to see if it showed differntial expression ( assuming differential promoter usage). If the tss_id was differentially expressed(statistically significantly) then I would assume the difference was because of promoter usage and not splicing per se. So I discarded the candidate. If tss_id wasn't differentially expressed I moved on to check the transcripts belonging to this tss_id in isoform_exp.diff file. Interesting candidates would be those transcripts that showed statistically significant differential splicing at this level. However, I barely found any candidates that matched my criteria. So am in a limbo.
I ran cufflinks using gtf as I was only interested in identifying the shift in isoform ratio of known transcripts due to alternative splicing. I do understand that one possible issue could be because of the presence of novel transcripts not reflected in the results of my pipeline. However I would really appreciate any insights or correction of my logic for finding validation candidates. A great help would be pointing at papers that have done so using similar approach/pipeline. I didn't find any so far.